Enhancing prediction accuracy for Parkinson's disease using advanced machine learning models | JoVE Visualize

Area of Science:

Neurology
Computer Science
Biomedical Engineering

Background:

Parkinson's disease (PD) is a progressive neurological disorder significantly impacting older adults' physical and mental health.
Current diagnostic methods for PD lack definitive treatments, highlighting the need for improved early detection strategies.
Machine learning (ML) offers a promising avenue for enhancing PD detection accuracy by analyzing complex datasets.

Purpose of the Study:

To develop and evaluate a descriptive machine learning (ML) approach for diagnosing Parkinson's disease (PD) using acoustical data.
To compare the efficacy of XGBoost, random forest (RF), support vector machine (SVM), and K-nearest neighbours (KNN) algorithms in classifying PD.
To assess the impact of data preprocessing techniques, including Synthetic Minority Over-sampling Technique (SMOTE) and Principal Component Analysis (PCA), on classification performance.

Main Methods:

Utilized the UCI Parkinson's disease dataset, focusing on acoustical features for analysis.
Implemented a four-phase research methodology: baseline model establishment, 10-fold cross-validation, class balancing with SMOTE, and dimensionality reduction with PCA.
Compared the performance of four common ML classification algorithms: XGBoost, RF, SVM, and KNN.

Main Results:

XGBoost achieved the highest classification accuracy (97.22%) and the best Matthews correlation coefficient, indicating its suitability for non-linear relationships in PD data.
SMOTE significantly improved the performance of ML algorithms, particularly KNN, while PCA effectively reduced dimensionality without information loss.
After applying SMOTE and PCA, the performance differences between ML algorithms diminished, suggesting statistically similar classifier performance.

Conclusions:

Machine learning algorithms, especially XGBoost and KNN, are effective tools for early Parkinson's disease detection using acoustical data.
The combination of SMOTE and PCA enhances the robustness and generalizability of ML models for PD diagnosis.
The integration of SHAP-based explainability ensures the interpretability of the ML framework, bridging the gap between computational predictions and clinical insights.